Prompt
We want one system for metrics, logs, and traces. Which time-series-adjacent database options are realistic?
Latest observation
If you want one backend for metrics, logs, and traces, the realistic “time-series-adjacent” options are usually not classic TSDBs alone, but observability stores that combine time-based ingestion with indexing/search and sometimes columnar analytics.
Realistic options
1) ClickHouse
Probably the most practical “single store” choice for many teams.
- Good for: logs, traces, metrics, analytics
- Why it works: columnar OLAP, very fast aggregations, compressed storage, can handle high ingest
- Strengths:
- Excellent for log search via structured fields
- Strong for trace/event analytics
- Can store metrics efficiently
- Mature, widely used in observability
- Tradeoffs:
- Not a native metrics TSDB with Prometheus-style query semantics
- Requires schema/design discipline
- Search is not like Elasticsearch full-text
- Typical ecosystem: Grafana, OpenTelemetry collectors, custom pipelines, SigNoz, Uptrace, Better Stack-style architectures
2) VictoriaMetrics
Best if your primary focus is metrics, with some adjacent log/tracing use only if you’re willing to compromise or pair it with other components.
- Good for: metrics first
- Strengths:
- Very efficient Prometheus-compatible storage
- Great compression and scale
- Simple operationally
- Tradeoffs:
- Not a general logs/traces backend
- Not realistic as a full single-store for all three unless your logs/traces requirements are minimal or transformed heavily
3) Apache Druid
More realistic for event analytics than for general logs/traces, but can sometimes be used in observability pipelines.
- Good for: high-ingest analytics, time-based event queries
- Strengths:
- Fast slice-and-dice over time-partitioned data
- Good for dashboards
- Tradeoffs:
- Not ideal for trace storage/query patterns
- Logs are possible, but not usually the best fit
- Operational complexity is higher
4) Apache Pinot
Similar to Druid, and often a better fit for real-time analytics workloads.
- Good for: low-latency analytics, event data, some observability use cases
- Strengths:
- Real-time ingestion
- Good OLAP query performance
- Tradeoffs:
- Not a native logs/traces system
- You’ll need a thoughtful schema and pipeline
- Less turnkey for observability than ClickHouse-based stacks
5) Elastic / OpenSearch
These are realistic if logs are the primary use case and metrics/traces are secondary.
- Good for: logs, search, text-heavy queries
- Strengths:
- Excellent full-text search and filtering
- Common in log-centric observability
- Tradeoffs:
- Metrics storage is workable but not ideal
- Traces can be stored, but it’s not the most efficient or elegant
- Cost and operational overhead can be significant at scale
6) Loki + Mimir/Prometheus + Tempo
Not one database, but worth mentioning because it’s the most common “observability trio.”
- Good for: best-of-breed approach
- Strengths:
- Very good fit for each signal type
- Grafana-native and popular
- Tradeoffs:
- Not one system
- More moving parts
- If your “one system” means “one platform,” this is often the pragmatic answer.
Less realistic as a single backend
These are usually not the right choice if you truly want metrics + logs + traces in one store:
- InfluxDB: great for metrics, not logs/traces
- Prometheus: metrics only
- TimescaleDB: strong time-series SQL, but logs/traces at scale are awkward
- QuestDB: strong time-series ingestion, but not a full observability backend
- IoT-focused TSDBs: often poor fit for logs/traces
- Elasticsearch alone: workable, but not ideal for metrics/traces efficiency
Practical ranking
If you want a realistic shortlist:
- ClickHouse — best all-around single-store candidate
- OpenSearch/Elasticsearch — best if logs/search dominate
- VictoriaMetrics — best if metrics dominate and “single store” is flexible
- Pinot/Druid — good analytics engines, less turnkey for observability
- TimescaleDB/InfluxDB/QuestDB — usually not enough for all three signals
Bottom line
If your goal is genuinely one database for metrics, logs, and traces, the most realistic option today is ClickHouse. It’s not a pure TSDB, but it’s the most credible single-store backend for unified observability workloads.
If you want, I can also give you:
- a decision matrix by workload size,
- a reference architecture for ClickHouse-based observability,
- or a comparison against Elasticsearch/OpenSearch and the Grafana stack.